nodes/maxomarai-ComfyUI-The-Last-Model-Switcher
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
Files
main
maxomarai-ComfyUI-The-Last-…/README.md
T
maxomaraiandClaude Opus 4.6 0b1bbd46ca Add first-class Ollama support for AI features 
Ollama is now an explicit provider option alongside anthropic/openai:
- Ollama (local LLM, free, no API key needed)
- Automatic 120s timeout for local providers (first request loads model)
- Helpful error messages with hints ("is Ollama running?")

AI Settings dialog rewritten:
- Clear provider descriptions with pricing/key info
- Skips API key prompt for local providers (ollama, custom)
- Model name examples per provider (llama3.2 for Ollama, gpt-4o for OpenAI)
- Prompts for base URL when using Ollama/custom
- Setup instructions shown after saving Ollama config

New button: "Test AI Connection"
- Sends minimal test prompt to verify setup
- Shows provider, model, and response
- Helpful error for connection failures (e.g. Ollama not running)

New backend endpoint: POST /ai_test
- Tests connection with "Reply with just: OK" prompt
- Uses longer timeout for local providers

README updated with Ollama setup guide.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-17 10:40:26 +02:00

7.3 KiB
Raw Permalink Blame History

The Last Model Switcher

A ComfyUI custom node by Maxomarai that makes switching between AI image generation models effortless.

Select your model from a dropdown and everything configures itself - CLIP, VAE, resolution, prompts, sampler settings, and guidance. Switch from SDXL to Flux without rewiring a single connection.

Quick Start

1. Install

cd ComfyUI/custom_nodes
git clone https://github.com/maxomarai/ComfyUI-The-Last-Model-Switcher.git the_last_model_switcher

Restart ComfyUI. The node appears under loaders > Maxomarai > The Last Model Switcher.

2. Connect

The node replaces multiple nodes in your workflow. Here's all you need:

The Last Model Switcher              KSampler
  ┌─────────────────────┐            ┌──────────────┐
  │ [model dropdown]    │            │              │
  │ [resolution]        │   MODEL ──>│ model        │
  │ [megapixels]        │            │              │
  │ [positive prompt]   │ positive ─>│ positive     │
  │ [negative prompt]   │ negative ─>│ negative     │
  │                     │            │              │
  │ width  [1024]       │     VAE ──>│         LATENT──> VAE Decode
  │ height [1024]       │            │              │
  │ steps  [20]         │   width ──>│              │
  │ cfg    [1.0]        │  height ──>│ EmptyLatent  │
  │ guidance [3.5]      │   steps ──>│ Image        │
  │                     │     cfg ──>│              │
  │ [info panel]        │            └──────────────┘
  └─────────────────────┘

That's it. No separate CLIPTextEncode, no FluxGuidance node, no guessing which VAE or text encoder to use.

3. Generate

Write your prompt, hit Queue. Switch models anytime - all settings update automatically.

What It Does

When you switch models... The node automatically...
SDXL to Flux 2 Changes CLIP, VAE, resolution, disables negative prompt, applies guidance
Flux 2 to SD 1.5 Switches to 512x512, enables negative prompt, adjusts CFG to 7.0
Any model change Updates steps, sampler, scheduler to optimal values for that model
Resolution preset change Updates width/height (keeps your steps/cfg untouched)
Megapixel change (Flux 2) Scales width/height while keeping aspect ratio

Outputs

Output Description
MODEL Loaded model (with ModelSamplingFlux for Flux)
CLIP Compatible text encoder (for custom encoding workflows)
VAE Compatible VAE
positive Your prompt encoded as conditioning (with FluxGuidance applied for Flux)
negative Negative prompt conditioning (empty for Flux - safe to keep connected)
width / height Resolution from preset, megapixel-scaled, or manually typed
steps / cfg Sampler settings (auto or manual)
guidance Flux guidance value (also applied internally to positive conditioning)

Features

Built-in Prompt Encoding

Write your positive and negative prompts directly on the node. They're encoded with the correct CLIP for the selected model. For Flux models, FluxGuidance is applied automatically to the positive conditioning.

Smart Warnings

The info panel warns you about connection issues in real-time:

  • Missing connections (MODEL, positive, negative not connected)
  • Empty prompt
  • Negative prompt on Flux (will be ignored)
  • Width/height not connected

Auto Model Scanner

Click "Scan for New Models" to detect models in your folders. The scanner reads safetensors headers (instant, no model loading) and identifies:

  • SD 1.5 / SDXL checkpoints
  • SD3 / SD3.5 models
  • Flux 1 Dev / Schnell
  • Flux 2 models
  • Turbo / Lightning / distilled variants (from filename)

Detected models get correct presets automatically.

AI Model Identification

Click "AI Identify Model" to use AI for precise model identification. It analyzes the filename, file size, and architecture to suggest optimal settings.

AI-Powered Features (multi-provider)

The node supports multiple AI providers for model identification and prompt enhancement:

Provider Cost API Key Notes
Anthropic (Claude) Paid Required Best quality for model identification
OpenAI (GPT) Paid Required Fast, widely available
Ollama Free Not needed Runs locally on your machine
Custom Varies Optional Any OpenAI-compatible endpoint (LM Studio, vLLM, etc.)

Click "AI Settings" to configure. Click "Test AI Connection" to verify.

Using Ollama (Local AI)

  1. Install from ollama.com and start it: ollama serve
  2. Pull a model: ollama pull llama3.2
  3. In the node, click "AI Settings" and choose ollama as provider
  4. Enter model name (e.g. llama3.2, qwen2.5, mistral)
  5. Click "Test AI Connection" to verify

Good local models for prompt work: llama3.2, qwen2.5:7b, mistral:7b.

Model Caching

Models are cached between executions. Re-running with the same model skips disk loading entirely. Cache clears automatically when you switch to a different model, freeing VRAM.

Live Value Sync

Connected downstream nodes (KSampler, EmptyLatentImage, etc.) update their widgets immediately when you change settings - before execution.

Supported Models (25 presets)

Only models you have downloaded appear in the dropdown.

Family Models
SD 1.5 Base, DreamShaper 8, Realistic Vision 6
SDXL Base, RealVisXL V5.0, Juggernaut XL, DreamShaper XL, Pony V6, Animagine XL, Playground v2.5, Lightning 4-Step, Turbo
SD3 / SD3.5 Large, Large fp8, Large Turbo, Medium
Flux 1 Dev, Dev GGUF Q5, Schnell, Kontext Dev GGUF
Flux 2 Dev (Mistral/Qwen3/full), Klein 4B fp8, Klein 9B

Adding Custom Models

Option A: Click "Scan for New Models" - auto-detects and adds presets.

Option B: Click "AI Identify Model" - uses AI for precise settings.

Option C: Click "Edit Presets File" - manually edit presets.json.

Node Buttons

Button What it does
Show Model Info Display current model details and recommended settings
Scan for New Models Auto-detect new models in your folders
AI Identify Model Use Claude AI for precise model identification
Reload Presets Reload presets.json after manual edits
Edit Presets File Show presets.json path (copied to clipboard)

Installation

Git Clone

cd ComfyUI/custom_nodes
git clone https://github.com/maxomarai/ComfyUI-The-Last-Model-Switcher.git the_last_model_switcher

Manual Download

  1. Download this repository as a ZIP
  2. Extract to ComfyUI/custom_nodes/the_last_model_switcher/
  3. Restart ComfyUI

Requirements

  • ComfyUI V2 (with comfy_api.latest support)
  • ComfyUI-GGUF (optional, for GGUF model support)
  • Anthropic API key (optional, for AI model identification)

License

MIT License - see LICENSE for details.

Made by Maxomarai

Reference in New Issue View Git Blame Copy Permalink